1. Field of the Invention
The present invention relates to an image processing apparatus and a target-region tracking program to track a target region indicating a tracking object that is shown in an object image (hereinafter, appropriately described as “frame”) at each time point obtained by capturing the tracking object at a plurality of time points in a time series.
2. Description of the Related Art
As a tracking processing of an object in images captured at a plurality of time points in a time series, Patent Document 1 discloses a method in which the object is tracked by associating target regions based on an overlapped area in serial frames of the target region indicating the tracking object, by categorizing a state of a target region as any one of appearance, single connection, coupling, disappearance, and separation based on a result of this association, and further by correcting, when categorized as coupling or separation, an error at the time of detection of the object. The error at the time of detection of the object is, for example, a phenomenon in which objects originally present in a plurality of pieces are detected as a single connected region due to overlap and the like, or in which an object originally being a single piece is detected as a plurality of separated regions due to some problems at the time of detection. Specific methods to correct such an error at the time of detection are following two methods.
In the first method, when regions indicating objects originally present in a plurality of pieces are connected into one region, it is determined as “connection”, and a feature quantity of each region before connection in a previous frame is recorded. Moreover, when the region determined as “connection” is separated, regions are associated with each other such that coincidence in a feature quantity becomes highest in various combinations of each region after separation and each region before connection, to perform tracking.
In the second method, when a region indicating an object originally being a single piece is separated, it is determined as “division”, and during the divided state, the tracking is performed assuming the center of gravity obtained by totaling respective regions being parts as a representative position. Moreover, when the regions determined as “division” are coupled, it is determined that two separated regions that are originally being a single piece has become detected as a single region.
To perform such correction, an object detection result with particularly high credibility is required in the serial frames. This is because even for separation, a different processing is performed for each of cases where a region that has been determined as connection is separated and where a region indicating an originally single object is separated, and also for coupling, a different processing is performed for each of cases where regions indicating originally plural objects are coupled and where regions that have been determined as division are coupled. In Japanese Patent Laid-Open Publication No. 2002-342762, to solve this problem, an initialization step is provided in which a result of detecting a region in a specific range in an image is used as a reference. This is because in Japanese Patent Laid-Open Publication No. 2002-342762, a tracking object is assumed to be a human, and for example, if an image range that shows around such a doorway that only one person can pass at a time is set as the specific range, it can be assumed that a result of object detection with high credibility can be obtained in that range.
However, in tracking a living cell, it is difficult to obtain an object detection result with high credibility in a specific range in an image. This is because in situations that a large number of dense cells are shot at a time, it is quite difficult to capture on a preferable condition independently for each of the cells and to detect a region showing the cell.